Download An Effective Visual Descriptor Based on Color and - MEX

LNAI 8856
Alexander Gelbukh
Félix Castro Espinoza
Sofía N. Galicia-Haro (Eds.)
Human-Inspired Computing
and Its Applications
13th Mexican International Conference
on Artificial Intelligence, MICAI 2014
Tuxtla Gutiérrez, Mexico, November 16–22, 2014
Proceedings, Part I
123
Lecture Notes in Artificial Intelligence
Subseries of Lecture Notes in Computer Science
LNAI Series Editors
Randy Goebel
University of Alberta, Edmonton, Canada
Yuzuru Tanaka
Hokkaido University, Sapporo, Japan
Wolfgang Wahlster
DFKI and Saarland University, Saarbrücken, Germany
LNAI Founding Series Editor
Joerg Siekmann
DFKI and Saarland University, Saarbrücken, Germany
8856
Alexander Gelbukh
Félix Castro Espinoza
Sofía N. Galicia-Haro (Eds.)
Human-Inspired Computing
and Its Applications
13th Mexican International Conference
on Artificial Intelligence, MICAI 2014
Tuxtla Gutiérrez, Mexico, November 16-22, 2014
Proceedings, Part I
13
Volume Editors
Alexander Gelbukh
Centro de Investigación en Computación
Instituto Politécnico Nacional
Mexico City, Mexico
E-mail: [email protected]
Félix Castro Espinoza
Universidad Autónoma del Estado de Hidalgo
Área Académica de Computación y Electrónica
Hidalgo, Mexico
E-mail: [email protected]
Sofía N. Galicia-Haro
Universidad Autónoma Nacional de México
Facultad de Ciencias
Mexico City, Mexico
E-mail: [email protected]
ISSN 0302-9743
e-ISSN 1611-3349
ISBN 978-3-319-13646-2
e-ISBN 978-3-319-13647-9
DOI 10.1007/978-3-319-13647-9
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2014955199
LNCS Sublibrary: SL 7 – Artificial Intelligence
© Springer International Publishing Switzerland 2014
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered and
executed on a computer system, for exclusive use by the purchaser of the work. Duplication of this publication
or parts thereof is permitted only under the provisions of the Copyright Law of the Publisher’s location,
in ist current version, and permission for use must always be obtained from Springer. Permissions for use
may be obtained through RightsLink at the Copyright Clearance Center. Violations are liable to prosecution
under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publication,
neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or
omissions that may be made. The publisher makes no warranty, express or implied, with respect to the
material contained herein.
Typesetting: Camera-ready by author, data conversion by Scientific Publishing Services, Chennai, India
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)
Preface
The Mexican International Conference on Artificial Intelligence (MICAI) is a
yearly international conference series organized by the Mexican Society of Artificial Intelligence (SMIA) since 2000. MICAI is a major international artificial
intelligence forum and the main event in the academic life of the country’s growing artificial intelligence community.
MICAI conferences publish high-quality papers in all areas of artificial intelligence and its applications. The proceedings of the previous MICAI events
have been published by Springer in its Lecture Notes in Artificial Intelligence
series, vols. 1793, 2313, 2972, 3789, 4293, 4827, 5317, 5845, 6437, 6438, 7094,
7095, 7629, 7630, 8265, and 8266. Since its foundation in 2000, the conference
has been growing in popularity and improving in quality.
According to two main areas of artificial intelligence—modeling human mental abilities on the one hand and optimization and classification on the other
hand—the proceedings of MICAI 2014 have been published in two volumes.
The first volume, Human-Inspired Computing and Its Applications, contains 44
papers structured into seven sections:
–
–
–
–
–
–
–
Natural Language Processing
Natural Language Processing Applications
Opinion Mining, Sentiment Analysis, and Social Network Applications
Computer Vision
Image Processing
Logic, Reasoning, and Multi-agent Systems
Intelligent Tutoring Systems
The second volume, Nature-Inspired Computation and Machine Learning,
contains 44 papers structured into eight sections:
–
–
–
–
–
–
–
–
Genetic and Evolutionary Algorithms
Neural Networks
Machine Learning
Machine Learning Applications to Audio and Text
Data Mining
Fuzzy Logic
Robotics, Planning, and Scheduling
Biomedical Applications
This two-volume set will be of interest to researchers in all areas of artificial
intelligence, students specializing in related topics, and to the general public
interested in recent developments in artificial intelligence.
The conference received for evaluation 350 submissions by 823 authors from
a record high number of 46 countries: Algeria, Argentina, Australia, Austria,
VI
Preface
Brazil, Bulgaria, Canada, Chile, China, Colombia, Cuba, Czech Republic,
Ecuador, Egypt, France, Germany, India, Iran, Ireland, Israel, Italy, Jordan,
Kazakhstan, Lithuania, Malaysia, Mexico, Morocco, Nepal, Norway, Pakistan,
Panama, Paraguay, Peru, Poland, Portugal, Russia, Singapore, Slovakia, South
Africa, Spain, Sweden, Turkey, UK, Ukraine, USA, and Virgin Islands (USA);
the distribution of papers by tracks is shown in Table 1. Of these submissions, 87
papers were selected for publication in these two volumes after a peer-reviewing
process carried out by the international Program Committee. The acceptance
rate was 24.8%.
In addition to regular papers, the second volume contains an invited paper
by Oscar Castillo, Patricia Melin, and Fevrier Valdez: “Nature-Inspired Optimization of Type-2 Fuzzy Systems.”
The international Program Committee consisted of 201 experts from 34 countries: Australia, Austria, Azerbaijan, Belgium, Brazil, Canada, China, Colombia,
Czech Republic, Denmark, Finland, France, Germany, Greece, India, Israel, Italy,
Japan, Mexico, The Netherlands, New Zealand, Norway, Poland, Portugal, Russia, Singapore, Slovenia, Spain, Sweden, Switzerland, Tunisia, Turkey, UK, and
USA.
Table 1. Distribution of papers by tracks
Track
Submissions
Natural Language Processing
59
Machine Learning and Pattern Recognition
42
Logic, Knowledge-Based Systems, Multi-Agent
40
Systems and Distributed AI
Computer Vision and Image Processing
38
Evolutionary and Nature-Inspired Metaheuristic
33
Algorithms
29
Data Mining
Neural Networks and Hybrid Intelligent Systems
28
Robotics, Planning and Scheduling
24
Fuzzy Systems and Probabilistic Models in
23
Decision Making
Bioinformatics and Medical Applications
18
Intelligent Tutoring Systems
16
Accepted
19
12
8
13
6
Rate
32%
29%
20%
34%
18%
7
7
5
24%
25%
21%
4
3
3
17%
17%
19%
MICAI 2014 was honored by the presence of such renowned experts as Hojjat
Adeli of The Ohio State University, USA, Oscar Castillo of Instituto Tecnológico
de Tijuana, Mexico, Bonnie E. John of IBM T.J. Watson Research Center, USA,
Bing Liu of the University of Illinois, USA, John Sowa of VivoMind Research,
USA, and Vladimir Vapnik of the NEC Laboratories, USA, who gave excellent
keynote lectures. The technical program of the conference also featured tutorials presented by Roman Bartak of Charles University, Czech Republic; Oscar
Castillo of Tijuana Institute of Technology, Mexico; Héctor G. Ceballos of Clark
& Parsia LLC, USA, and Héctor Pérez Urbina of Tecnológico de Monterrey,
Mexico; Sanjoy Das of Kansas State University, USA; Alexander Gelbukh of
Preface
VII
Instituto Politécnico Nacional, Mexico; Bonnie E. John of IBM T. J. Watson
Research Center, USA; Bing Liu of University of Illinois, USA; Raúl Monroy of
Tecnológico de Monterrey, Mexico; John Sowa of VivoMind Research, USA; and
Luis Martı́n Torres Treviño of Universidad Autónoma de Nuevo León, Mexico,
among others. Three workshops were held jointly with the conference: the 7th
International Workshop on Hybrid Intelligent Systems, HIS 2014; the 7th International Workshop on Intelligent Learning Environments, WILE 2014, and the
First International Workshop on Recognizing Textual Entailment and Question
Answering, RTE-QA 2014.
The authors of the following papers received the Best Paper Award on the
basis of the paper’s overall quality, significance, and originality of the reported
results:
1st place: “The Best Neural Network Architecture,” by Angel Kuri-Morales
(Mexico)
2nd place: “Multisensor-Based Obstacles Detection in Challenging Scenes,” by Yong
Fang, Cindy Cappelle, and Yassine Ruichek (France)
“Intelligent Control of Induction Motor-Based Comparative Study:
Analysis of Two Topologies,” by Moulay Rachid Douiri, El Batoul
Mabrouki, Ouissam Belghazi, Mohamed Ferfra, and Mohamed Cherkaoui
(Morocco)
3rd place: “A Fast Scheduling Algorithm for Detection and Localization of Hidden
Objects Based on Data Gathering in Wireless Sensor Networks,” by
Eugene Levner, Boris Kriheli, Amir Elalouf, and Dmitry Tsadikovich
(Israel)
The authors of the following papers selected among all papers of which the
first author was a full-time student, excluding the papers listed above, received
the Best Student Paper Award:
1st place: “Solving Binary Cutting Stock with Matheuristics,” by Ivan Adrian Lopez
Sanchez, Jaime Mora Vargas, Cipriano A. Santos, and Miguel Gonzalez
Mendoza (Mexico)
“Novel Unsupervised Features for Czech Multi-label Document
Classification,” by Tomáš Brychcín and Pavel Král (Czech Republic)
We want to thank everyone involved in the organization of this conference.
In the first place, these are the authors of the papers published in this book: It is
their research work that gives value to the book and to the work of the organizers.
We thank the track chairs for their hard work, the Program Committee members,
and additional reviewers for their great effort spent on reviewing the submissions.
We are grateful to the Dean of the Instituto Tecnológico de Tuxtla Gutiérrez (ITTG), M.E.H. José Luis Méndez Navarro, the Dean of the Universidad
Autónoma de Chiapas (UNACH), Professor Jaime Valls Esponda, and M.C.
Francisco de Jesús Suárez Ruiz, Head of IT Department, for their instrumental support of MICAI and for providing the infrastructure for the keynote talks,
VIII
Preface
tutorials, and workshops, and to all professors of the Engineering School of Computational Systems for their warm hospitality and hard work, as well as for their
active participation in the organization of this conference. We greatly appreciate the generous sponsorship provided by the Government of Chiapas via the
Conventions and Visitors Office (OCV).
We are deeply grateful to the conference staff and to all members of the Local
Committee headed by Imelda Valles López. We gratefully acknowledge support
received from the project WIQ-EI (Web Information Quality Evaluation Initiative, European project 269180). The entire submission, reviewing, and selection
process, as well as preparation of the proceedings, was supported for free by the
EasyChair system (www.easychair.org). Finally, yet importantly, we are very
grateful to the staff at Springer for their patience and help in the preparation of
this volume.
October 2014
Alexander Gelbukh
Félix Castro Espinoza
Sofı́a N. Galicia-Haro
Table of Contents – Part I
Natural Language Processing
Finding the Most Frequent Sense of a Word by the Length of Its
Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hiram Calvo and Alexander Gelbukh
1
Complete Syntactic N-grams as Style Markers for Authorship
Attribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Juan-Pablo Posadas-Duran, Grigori Sidorov, and Ildar Batyrshin
9
Extracting Frame-Like Structures from Google Books NGram
Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Vladimir Ivanov
18
Modeling Natural Language Metaphors with an Answer Set
Programming Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Juan Carlos Acosta-Guadarrama, Rogelio Dávila-Pérez,
Mauricio Osorio, and Victor Hugo Zaldivar
Whole-Part Relations Rule-Based Automatic Identification: Issues from
Fine-Grained Error Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ilia Markov, Nuno Mamede, and Jorge Baptista
28
37
Statistical Recognition of References in Czech Court Decisions . . . . . . . . .
Vincent Krı́ž, Barbora Hladká, Jan Dědek, and Martin Nečaský
51
LSA Based Approach to Domain Detection . . . . . . . . . . . . . . . . . . . . . . . . . .
Diego Uribe
62
Natural Language Processing Applications
Best Student Paper Award
Novel Unsupervised Features for Czech Multi-label Document
Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Tomáš Brychcı́n and Pavel Král
Feature Selection Based on Sampling and C4.5 Algorithm to Improve
the Quality of Text Classification Using Naı̈ve Bayes . . . . . . . . . . . . . . . . .
Viviana Molano, Carlos Cobos, Martha Mendoza,
Enrique Herrera-Viedma, and Milos Manic
70
80
XX
Table of Contents – Part I
Detailed Description of the Development of a MOOC in the Topic
of Statistical Machine Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Marta Ruiz Costa-jussà, Lluı́s Formiga, Jordi Petit,
and José A.R. Fonollosa
NEBEL: Never-Ending Bilingual Equivalent Learner . . . . . . . . . . . . . . . . . .
Thiago Lima Vieira and Helena de Medeiros Caseli
92
99
RI for IR: Capturing Term Contexts Using Random Indexing
for Comprehensive Information Retrieval . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Rajendra Prasath, Sudeshna Sarkar, and Philip O’Reilly
104
Data Extraction Using NLP Techniques and Its Transformation to
Linked Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Vincent Krı́ž, Barbora Hladká, Martin Nečaský, and Tomáš Knap
113
A New Memetic Algorithm for Multi-document Summarization Based
on CHC Algorithm and Greedy Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Martha Mendoza, Carlos Cobos, Elizabeth León, Manuel Lozano,
Francisco Rodrı́guez, and Enrique Herrera-Viedma
125
How Predictive Is Tense for Language Profiency? A Cautionary Tale . . .
Alexandra Panagiotopoulos and Sabine Bergler
139
Evaluating Term-Expansion for Unsupervised Image Annotation . . . . . . .
Luis Pellegrin, Hugo Jair Escalante, and Manuel Montes-y-Gómez
151
Opinion Mining, Sentiment Analysis, and Social
Network Applications
Gender Differences in Deceivers Writing Style . . . . . . . . . . . . . . . . . . . . . . .
Verónica Pérez-Rosas and Rada Mihalcea
163
Extraction of Semantic Relations from Opinion Reviews in Spanish . . . . .
Sofı́a N. Galicia-Haro and Alexander Gelbukh
175
Evaluating Polarity for Verbal Phraseological Units . . . . . . . . . . . . . . . . . . .
Belém Priego Sánchez, David Pinto, and Salah Mejri
191
Restaurant Information Extraction (Including Opinion Mining
Elements) for the Recommendation System. . . . . . . . . . . . . . . . . . . . . . . . . .
Ekaterina Pronoza, Elena Yagunova, Svetlana Volskaya,
and Andrey Lyashin
Aggressive Text Detection for Cyberbullying . . . . . . . . . . . . . . . . . . . . . . . .
Laura P. Del Bosque and Sara Elena Garza
201
221
Table of Contents – Part I
A Sentiment Analysis Model: To Process Subjective Social Corpus
through the Adaptation of an Affective Semantic Lexicon . . . . . . . . . . . . .
Guadalupe Gutiérrez, Lourdes Margain, Carlos de Luna,
Alejandro Padilla, Julio Ponce, Juana Canul, and Alberto Ochoa
Towards Automatic Detection of User Influence in Twitter by Means
of Stylistic and Behavioral Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Gabriela Ramı́rez-de-la-Rosa, Esaú Villatoro-Tello,
Héctor Jiménez-Salazar, and Christian Sánchez-Sánchez
XXI
233
245
Computer Vision
Best Paper Award, Second Place
Multisensor Based Obstacles Detection in Challenging Scenes . . . . . . . . . .
Yong Fang, Cindy Cappelle, and Yassine Ruichek
257
GP-MPU Method for Implicit Surface Reconstruction . . . . . . . . . . . . . . . .
Manuel Guillermo López, Boris Mederos, and Oscar S. Dalmau
269
Monocular Visual Odometry Based Navigation for a Differential Mobile
Robot with Android OS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Carla Villanueva-Escudero, Juan Villegas-Cortez,
Arturo Zúñiga-López, and Carlos Avilés-Cruz
Comparison and Analysis of Models to Predict the Motion of Segmented
Regions by Optical Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Angel Juan Sanchez Garcia, Maria de Lourdes Velasco Vazquez,
Homero Vladimir Rios Figueroa, Antonio Marin Hernandez,
and Gerardo Contreras Vega
Image Based Place Recognition and Lidar Validation for Vehicle
Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Yongliang Qiao, Cindy Cappelle, and Yassine Ruichek
281
293
304
Image Processing
Frequency Filter Bank for Enhancing Carbon Nanotube Images . . . . . . . .
José de Jesús Guerrero Casas, Oscar S. Dalmau, Teresa E. Alarcón,
and Adalberto Zamudio
A Supervised Segmentation Algorithm for Crop Classification Based on
Histograms Using Satellite Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Francisco E. Oliva, Oscar S. Dalmau, and Teresa E. Alarcón
316
327
XXII
Table of Contents – Part I
An Effective Visual Descriptor Based on Color and Shape Features for
Image Retrieval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Atoany Fierro-Radilla, Karina Perez-Daniel,
Mariko Nakano-Miyatakea, Hector Perez-Meana,
and Jenny Benois-Pineau
Compressive Sensing Architecture for Gray Scale Images . . . . . . . . . . . . . .
Gustavo Gonzalez-Garcia, Alfonso Fernandez-Vazquez,
and Rodolfo Romero-Herrera
A Novel Approach for Face Authentication Using Speeded Up Robust
Features Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cyntia Mendoza-Martinez, Jesus Carlos Pedraza-Ortega,
and Juan Manuel Ramos-Arreguin
On-Line Dense Point Cloud Generation from Monocular Images with
Scale Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ander Larranaga-Cepeda, Jose Gabriel Ramirez-Torres,
and Carlos Alberto Motta-Avila
336
349
356
368
An Improved Colorimetric Invariants and RGB-Depth-Based Codebook
Model for Background Subtraction Using Kinect . . . . . . . . . . . . . . . . . . . . .
Julian Murgia, Cyril Meurie, and Yassine Ruichek
380
Novel Binarization Method for Enhancing Ancient and Historical
Manuscript Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Saad M. Ismail and Siti Norul Huda Sheikh Abdullah
393
Logic, Reasoning, and Multi-agent Systems
Preferences for Argumentation Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mauricio Osorio, Claudia Zepeda, and José Luis Carballido
Computing Preferred Semantics: Comparing Two ASP Approaches vs
an Approach Based on 0-1 Integer Programming . . . . . . . . . . . . . . . . . . . . .
Mauricio Osorio, Juan Dı́az, and Alejandro Santoyo
Experimenting with SAT Solvers in Vampire . . . . . . . . . . . . . . . . . . . . . . . .
Armin Biere, Ioan Dragan, Laura Kovács, and Andrei Voronkov
Onto Design Graphics (ODG): A Graphical Notation to Standardize
Ontology Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Rafaela Blanca Silva-López, Mónica Silva-López,
Iris Iddaly Méndez-Gurrola, and Maricela Bravo
A Logic for Context-Aware Non-monotonic Reasoning Agents . . . . . . . . .
Abdur Rakib and Hafiz Mahfooz Ul Haque
407
419
431
443
453
An Effective Visual Descriptor
Based on Color and Shape Features for Image Retrieval
Atoany Fierro-Radilla1, Karina Perez-Daniel1, Mariko Nakano-Miyatakea1,
Hector Perez-Meana1, and Jenny Benois-Pineau2
1
Postgraduate Section of ESIME Culhuacan, Instituto Politecnico Nacional,
Av. Santa Ana No. 1000, Col. San Francisco Culhuacan
[email protected], [email protected]
2
University of Bordeaux I, Bordeaux, France
[email protected]
Abstract. In this paper we present a Content-Based Image Retrieval (CBIR)
system which extracts color features using Dominant Color Correlogram Descriptor (DCCD) and shape features using Pyramid Histogram of Oriented Gradients (PHOG). The DCCD is a descriptor which extracts global and local color
features, whereas the PHOG descriptor extracts spatial information of shape in
the image. In order to evaluate the image retrieval effectiveness of the proposed
scheme, we used some metrics commonly used in the image retrieval task such
as, the Average Retrieval Precision (ARP), the Average Retrieval Rate (ARR)
and the Average Normalized Modified Retrieval Rank (ANMRR) and the Average Recall (R)-Average Precision (P) curve. The performance of the proposed
algorithm is compared with some other methods which combine more than one
visual feature (color, texture, shape). The results show a better performance of
the proposed method compared with other methods previously reported in the
literature.
Keywords: CBIR, color descriptor, shape descriptor, dominant color, color correlogram, PHOG.
1
Introduction
In the last years, due to the technological advances, a large amount of devices such as:
digital cameras, smart phones and tablets, have been developed in order to capture
images and video data and on the other hand, a technological advance on high-speed
internet connection, as well as the increasing storage capacities, leads to a growing
size of databases. As a result Internet has become the largest multimedia database; a
huge amount of information becomes available for a large number of users [1]. With
large databases, it is a challenge to browse and retrieve efficiently the desirable information. The traditional annotation heavily relies on manual labor to label images
with keywords, which unfortunately can hardly describe the diversity and ambiguity
for image contents [2].
The Content-Based Image Retrieval (CBIR) system is a useful tool to resolve the
above mentioned problem. The typical CBIR system performs two major tasks, the
A. Gelbukh et al. (Eds.): MICAI 2014, Part I, LNAI 8856, pp. 336–348, 2014.
© Springer International Publishing Switzerland 2014
An Effective Visual Descriptor Based on Color and Shape Features
337
first one is the feature extraction, where a set of features is extracted to describe the
content of each image in the database; and the second task is the similarity measurement between the query image and each image in the database, using the extracted
features [3]. Generally the CBIR is performed using some low-level visual descriptors
such as color-based, texture-based and shape-based descriptors, which extract feature
vectors from the images.
There are many methods which combine more than one visual descriptor [2-6],
improving the image retrieval effectiveness. In [2], authors combine Linear Block
Algorithm (LBA) for the global color feature extraction, Steerable Filter for the texture features extraction and Pseudo-Zernike Moments for extraction of the shape features which are rotation invariant. Authors of [3] combine color and texture features,
in which the color features are extracted using the Color Layout Descriptor (CLD)
and the texture features are obtained using the Gabor Filters. Another method that
combines more than one visual descriptors is proposed in [6], in which the image is
divided into six blocks then, the color space of each block is converted from RGB to
HSV and a cumulative histogram is computed in order to obtain the color features,
whereas to obtain the texture features of each block, four statistic features, such as
energy, contrast, entropy and inverse difference from the Gray-Level Concurrence
Matrix (GLCM), are computed.
In this paper we propose a method that combines local and global color information using the Dominant Color Correlogram Descriptor (DCCD) [7] and local shape
information using the Pyramid Histogram of Oriented Gradients (PHOG) [8]. The
proposed scheme is performed through three stages: In the first stage, the algorithm
obtains global color features using the Dominant Color Descriptor (DCD) proposed
by MPEG-7 [9, 10] as well as the shape information using the PHOG [10]. In the
second stage, using color correlogram the correlation between central pixel and its
neighborhood is calculated from the image represented by only dominant colors, and
in the third stage, color and shape features are combined in order to obtain a new visual descriptor which improves the image retrieval performance. In order to evaluate
the proposed visual descriptor, we use three metrics commonly used in the CBIR
systems, such as ARP (Average Retrieval Precision), ARR (Average Retrieval Rank)
and ANMRR (Average Normalized Retrieval Rank), as well as RP-curves. The performance of our proposed scheme is compared with some methods reported in the
literature and the results show that the proposed visual descriptor improves the image
retrieval performance.
The rest of this paper is organized as follows: In Section 2 we briefly describe
some color-based descriptors commonly used in the literature. In Section 3, we briefly
describe some shape-based descriptors reported in the literature. In Section 4, we
present the proposed scheme. The results are shown in Section 5 and finally in Section 6 we present the conclusions of this work.
2
Color-Based Descriptors
Color is the basic element of image content and one of the main sensation features
when a human distinguish images [9]. From the perspective of feature extraction,
338
A. Fierro-Radilla et al.
color-based image descriptor can be divided into two categories [12]: Global descriptor which takes into account the whole image in order to obtain the color features. In
this group we have Histogram Intersection (HI) and Dominant Color Descriptor
(DCD). On the other hand we have the local descriptors such as Color Correlogram
(CC), Color Layout Descriptor (CLD) and Color Structure Descriptor (CSD), which
obtain the color features by dividing the image into regions.
The Histogram Intersection (HI) was proposed by Swain Ballard [13]. This method
is a global color descriptor and it is defined as: Given a pair of color histograms, I and
M, with n bins each one, HI can be computed as:
∑
,
min
,
(1)
This method is robust to geometrical modification, such as rotation and scaling, as
well as the variation of the image resolution [14]. The number of bins is an important
factor, because the more bins are used, the image is better described but, the computational cost is increased. Another global color descriptor is the DCD, which was proposed in the MPEG-7 standard. This color descriptor replaces the whole image color
information with a small number of representative colors [9]. The Dominant Color
Descriptor can be defined as follows:
,
,
1,2, … , ,
0,1 ,
(2)
where
is the percentage of the dominant color .
The color correlogram (CC) is a local color descriptor which expresses how the
spatial correlation of pairs of color changes with the distance [15] and it is defined as:
For any pixel of color
in the image, the color correlogram (
) gives the probability that a pixel at distance away from the given pixel ci has a color .The Color
Layout Descriptor (CLD) is a compact descriptor which represents the color spatial
distribution of visual data [16], where the color space used in this descriptor is the
YCbCr. The extraction of this descriptor consists of four stages. In the first one the
image is partitioned into 64 blocks where the size of each block is W/8 and H/8 with
W and H denoting the width and height of the image. In the second stage, for each
block, a single dominant color is selected, then in the third stage the three components
of the color space are transformed into 8x8 DCT (Discrete Cosine Transform). And
finally in the fourth stage, the DCT coefficients of Y, Cb and Cr color channel are
quantized and their lower coefficients are extracted to form the CLD. Another local
color descriptor is the Color Structure Descriptor, which was also proposed in the
MPEG-7 standard. In which an image is represented by the color distribution, and the
local spatial structure of color using structuring element. It is similar to color histogram but it is semantically different. The CSD is defined as
,
1, … , ,
where the bin value
is the number of structuring elements containing one or
more pixels with color
. Denote I be the set of quantized index of an image and
be the set of quantized color index existing inside the sub-image region covered
by the structuring element [17], the color histogram bins are accumulated according to
1,
(3)
An Effective Visual Descriptor Based on Color and Shape Features
3
339
Shape-Based Descriptor
Shape feature is an important factor in order to identify objects as well as classification and indexing of the context semantically. In this section, three shape-based descriptors, which are Pseudo-Zernike Moments (PZM), Polar Harmonic Transform
(PHT) and Pyramid Histogram of Oriented Gradients (PHOG), are described.
3.1
Pseudo-Zernike Moment (PZM)
The PZM consists of a set of complex polynomials that form a complete orthogonal
set over the interior of the unit circle,
1 [2]. These polynomials are denoted as
,
,
,
(4)
is the distance from the origin to the pixel ,
and is an
where
angle between vector and the -axis in the clockwise direction. The radial polyis defined as:
nomial
The PZM of order
!
| |
∑
| |
!
with repetition
!
| |
(5)
!
is defined as:
,
,
(6)
For a digital image of size MxN, its PZM can be computed as:
∑
∑
,
,
(7)
where ∆
,∆
The integer numbers and
are defined in Table 1. A numerical instability,
when high-order PZMs is required, is a serious problem of the PZM, due to the
amount of factorial elements in the radial polynomial.
Table 1. Principal Pseudo-Zernike Moments
Order
Moments
No. Moments
Order
Moments
No. Moments
0
1
4
,,
,
3
1
1
5
,
,
3
6
2
,
2
3
,
2
,
,
,
4
340
3.2
A. Fierro-Radilla et al.
Polar Harmonic Transform (PHT)
The Polar Complex Exponential Transform (PCET) [18] is one of the PHT, which is
defined as (8), when the order is and the repetition is , | | | | 0,1, … , ∞.
,
,
(8)
where
,
(9)
The radial kernel is a complex exponential in the radial direction, that is
(10)
For a digital image of size MxN, the PCET can be computed as:
∑
where ∆
,∆
∑
,
∆ ∆
(11)
,
(12)
, we finally obtain:
∑
3.3
,
∑
,
Pyramidal Histogram of Oriented Gradients (PHOG)
The Pyramidal Histogram of Oriented Gradients (PHOG) is a spatial shape descriptor
which represents the spatial distribution of edges and it is formulated as a vector representation [8]. The operation of PHOG consists of the following four steps.
1. Edge contour extraction: The contour of input image can be extracted using the
Canny edge detector.
2. Cell division: The edge detected binary image is divided into cells at several pyramid levels. For example, in the first pyramid level, the edge image is divided into
2×2 cells and in the second pyramid level, each cell furthermore is divided into
2×2 sub-cells. The cell division is repeated until desirable resolution levels of pyramid.
3. HOG calculation: The Histogram of Oriented Gradients (HOG) of each cell is calculated at each pyramid resolution level. The HOG of each cell in the same pyramid level is concatenated to form a vector.
4. PHOG extraction: The final PHOG is a concatenation of all HOG vectors generates
in all pyramid levels.
An Effective Visual Descriptor Based on Color and Shape Features
341
Fig. 1. Block diagram of the proposed scheme
4
Proposed Visual Descriptor
In this paper we propose a new visual descriptor which is a combination of the global
and local color information based on the DCCD and the shape feature based on the
PHOG. The block diagram of the proposed method is shown in fig. 1.
4.1
Global Color Information Extraction
In this paper we selected the HSV (Hue, Saturation, Value) color space because it is
similar in the manner that human distinguish the colors [7]. In order to reduce computational cost, the quantization in the HSV color space is performed using a nonlinear quantization, which is given by
0
1
2
3
316,20
20,40
40,75
75,155
4
5
6
7
155,190
190,270
270,295
295,316
(13)
342
A. Fierro-Radilla et al.
0
0,0.2
1
2
0.2,0.7 ,
0.7,
0
1
0,0.2
0.2,0.7
2
0.7,
(14)
The three quantized components are then combined into one matrix:
9
3
(15)
As a result, it is obtained a matrix with only 8x3x3=72 colors. From the matrix calculated in (15) the dominant colors are extracted using the DCD operation [9,10].
4.2
Local Color Information and Shape Extraction
The color correlogram is computed using the structuring element of 3x3 pixels which
scans the image, calculating the correlation between the central pixel and its neighborhood at a distance k=1. Thus, the proposed scheme gives spatial information. The
color correlogram can be computed as:
1
, 2
|
2
1
2
1
(16)
where
is i-th color and
and
are any two pixels in the input image I.
In the proposed scheme, we used auto-color correlogram, in which two pixels have
the same the color. As a result we obtain a color-based descriptor called the Dominant
Color Correlogram Descriptor (DCCD) and is defined as:
,
(17)
where
is the color auto-correlogram of the i-th dominant color .
The shape feature extraction is done using the PHOG descriptor mentioned in 3.3,
in which we used 3 pyramid levels and 8 bins for computing the HOG in each level.
4.3
Combination of Features
In the literature there are many manners to combine more than one visual feature, in
this paper we used the linear combination method used in [16], which is as follows.
,
(18)
1
(19)
where and are color and shape distances between the query and dataset image,
respectively. Firstly the two distances are normalized by (18) and combined lineally
by (19) to obtain the similarity Q, where
is the weight of a particular visual feature. If this similarity is smaller, two images are considered as more similar.
An Effective Visual Descriptor Based on Color and Shape Features
343
Table 2. Performance of color-based descriptors using dataset 1
Method
ANMRR
ARR
ARP
CC
[15]
HI
[13]
DCD
[9]
LBA
[2,10]
CLD
[3,16, 21]
CSD
[17]
DCCD
[7]
0.3126
α=2
0.7577
α=1
0.5923
α=1
0.5923
α=0.5
0.7846
α=0.25
0.8923
0.2507
0.8115
0.6269
0.6269
0.7923
0.9077
0.2576
0.8154
0.6154
0.6154
0.7846
0.8615
0.3579
0.6808
0.5642
0.5642
0.7154
0.8154
0.3358
0.7385
0.5731
0.5731
0.7154
0.8000
0.3145
0.7538
0.5846
0.5846
0.7538
0.8769
0.2266
0.8231
0.6808
0.6808
0.8538
0.9231
Table 3. Performance of color-based descriptors using dataset 2
Method
CC
[15]
HI
[13]
DCD
[9]
LBA
[2,10]
CLD
[3,16, 21]
CSD
[17]
DCCD
[7]
ANMRR
ARR
ARP
α=1
0.5870
α=1
0.5870
α=0.5
α=0.25
0.3228
α=2
0.7200
0.7620
0.8880
0.3174
0.7610
0.5760
0.5760
0.7380
0.8640
0.3384
0.7420
0.5590
0.5590
0.6920
0.8480
0.3478
0.7320
0.5500
0.5500
0.7040
0.8000
0.3194
0.7620
0.5740
0.5740
0.7280
0.8360
0.4431
0.6190
0.4630
0.4630
0.6200
0.7680
0.3086
0.7590
0.5960
0.5960
0.7560
0.8840
344
5
A. Fierro-Radilla et al.
a
Experimental Results
R
As we mentioned before, there
t
are many visual descriptors, such as color-based, ttexture-based and shape-based
d, also, there are many methods which combine these descriptors in order to improv
ve the image retrieval performance. We analyzed and eevaluate some algorithms using
g three different datasets; the dataset 1 is composed by 5500
images, divided into 25 caategories with 20 ground truth images per category. T
The
dataset 2 is composed by 1000 images, divided into 20 categories with 50 grouund
truth images per category, and
a the dataset 3 is called Corel Dataset 1k [19, 20] whhich
is composed by 1000 imagees, divided into 10 categories with 100 ground truth imaages
per category. The first two datasets (dataset 1 and dataset 2) are randomly selectedd by
authors, considering differeent percentage of query images respect to the dataset ssize
and the third one is commo
only used in CBIR evaluation. The color-based descripttors
were evaluated using dataset 1 and dataset 2 and the results are shown in table 2 and
table 3.
In the table 2 and 3, we can observe that, the results obtained by the DCCD, pproposed in [7], are better than
n the others. The RP-curves, shown in figures 2 and 3, can
describe the performance off the descriptors in the image retrieval task:
Fig. 2. RP-curve using dataset 1
An Effectiv
ve Visual Descriptor Based on Color and Shape Features
345
Fig. 3. RP-curve using dataset 2
The shape-based shape--based descriptors, PZM, PCET, PCT and PHOG, are eevaluated. The evaluation was done using dataset 1, and the results are shown in the following table:
Table 4. Performance of shape-based descriptors using dataset 1
Method
PZM
[2]
PCET
[18]
PCT
[18]
PHOG
[8]
ANMRR
ARR
ARP
0.7177
α=2
α
0
0.3808
α=1
0.2000
α=1
0.2000
α=0.5
0.2308
α=0.25
0.3692
0.7212
0
0.3642
0.2231
0.2231
0.2615
0.3692
0.7066
0
0.3554
0.2423
0.2423
0.2538
0.3692
0.5825
0
0.4462
0.3538
0.3538
0.4692
0.6308
From tables 2, 3 and 4 and
a figures 2 and 3, we can observe that the DCCD and the
PHOG perform better than the other descriptors, so we decided to combine the DC
CCD
color-based descriptor and the PHOG shape-based descriptor in this paper. We uused
the combination of two feaature vectors given by (19) with three different weighht ω
[16] and a simple concatenaation of two feature vectors.
346
A. Fierro-Radilla et al.
Table 5. Combination of two feature vectors with different weght ω using dataset 1
Method
Q
ω=0.8
Q
ω=0.5
Q
ω=0.3
ANMRR
ARR
ARP
0.3933
α=2
0.6385
α=1
0.5077
α=1
0.5077
α=0.5
0.7000
α=0.25
0.8000
0.2364
0.8077
0.6769
0.6769
0.8154
0.9231
0.2150
0.8309
0.6846
0.6846
0.8615
0.9538
0.2227
0.8231
0.6808
0.6808
0.8538
0.9385
Concatenation
The table 5 shows that using weight
0.3, the performance of image retrieval
task is better, which means the contribution of color feature is more important than
the shape feature in the CBIR. As we mentioned before, the proposed scheme combines color and shape features based on the DCCD and the PHOG using the combination method given by (19) to obtain the similarity Q. We compared the proposed
scheme with the method proposed in [2], which combines color, texture and shape
features, using the dataset 1 and dataset 2. In the evaluation, we employed three metrics commonly used in the CBIR, such as Average Normalized Retrieval Rank
(ANMRR), Average Retrieval Rate (ARR) and Average Retrieval Precision (ARP)
[7, 14]. The number of queries must be at least 1% of the dataset size [12]. For dataset
1, we used 13 queries for evaluation equivalent to the 2.6% and for dataset 2 we used
20 queries equivalent to the 2% of the dataset size. The comparison results between
the proposed scheme and [2] are shown in table 6.
Table 6. Comparison with the previous method [2]
dataset 1
Method
[2]
Proposed
dataset 2
[2]
Proposed
ANMRR
ARR
ARP
0.3425
α=2
0.7308
α=1
0.5346
α=1
0.5346
α=0.5
0.7308
α=0.25
0.8615
0.2150
0.8309
0.6846
0.6846
0.8615
0.9538
0.3672
0.6750
0.5420
0.5420
0.7020
0.8400
0.2698
0.7800
0.6550
0.6550
0.8120
0.9320
Using Corel Dataset 1k, we compare the proposed algorithm with the algorithms
proposed in [3-6]. The evaluation method used here is the same one used in [3], in
which 80 queries are used corresponding to the 8% of the dataset size. The query
images are the same used in [3] and the number of retrieved images to compute the
An Effective Visual Descriptor Based on Color and Shape Features
347
Average Precision (AP) is the same as well. The results are shown in table 7. From
the table, we can conclude the proposed method globally outperforms four previously
proposed methods [3-6].
Table 7. Comparison with several methods [3-6]
6
Class name
[3]
[4]
[5]
[6]
Proposed
Tribe
54 %
44.1%
32.3%
41%
61%
Beach
38%
30.6%
61.2%
32%
46.6%
Buildings
40%
38.2%
39.2%
37%
41.09%
Buses
64%
67.6%
39.5%
66%
62.93%
Dinosaurs
96%
97.2%
99.6%
43%
93.52%
Elephants
62%
33.8%
55.7%
39%
41.78%
Roses
68%
88.8%
89.3%
87%
77.51%
Horses
75%
63.2%
65.2%
35%
77.35%
Mountains
45%
31.3%
56.8%
34%
42.79%
Food
53%
34.9%
44.1%
31%
68.59%
Average
59.5%
52.97%
58.29%
44.5%
61.26%
Conclusions
In this paper we proposed a scheme which combines color and shape features for the
content-based image retrieval (CBIR) task. The proposed scheme extracts both global
and local color information using the DCCD and the shape feature based on boundary
information of objects using the PHOG. Two features are combined by weighted linear combination. We set a greater color-weight because the color feature provides
the most distinguishable information compared with the shape features. The comparison of the proposed algorithm with four previously proposed algorithms, which combine more than one visual descriptor, shows the better performance of the proposed
algorithm.
Acknowledgements. We thanks to the National Council of Science and Technology
of Mexico (CONACyT), to ANR of France and the National Polytechnic Institute for
the financial support during the realization of this research.
348
A. Fierro-Radilla et al.
References
1. Penatti, O., Valle, E., Torres, S.: Comparative study of global color and texture descriptors
for web image retrieval. J. Vis. Comm. Image Representation 23, 359–380 (2012)
2. Wang, X.Y., Yu, Y.J., Yang, H.Y.: An effective image retrieval scheme using color,
texture and shape features. Computer Standards & Interfaces 33, 59–68 (2011)
3. Jalab, H.A.: Image retrieval system based on color layout descriptor and global filters. In:
IEEE Conf. on Open System, pp. 32–36 (2011)
4. Pujari, J., Hiremath, P.: Content-based image retrieval based on color, texture and shape
features. Signal and Image Processing, 239–242 (2010)
5. Hafiane, A., Zavidovique, B.: Local relational string and mutual matching doe image
retrieval. Information Processing &Manegement 44, 1201–1212 (2008)
6. Kavitha, C., Prabhakara, B., Govardhan, A.: Image retrieval based on color and texture
features of the image subblocks. Int. J. Comput. Appl. 15, 33–37 (2011)
7. Fierro, A., Perez, K., Nakano, M., Benois, J.: Dominant color correlogram descriptor for
content-based image retrieval. In: 3rd Int. Conf. on Image, Vision and Computing (2014)
8. Yang, B., Guo, L., Jin, L., Huang, Q.: A novel feature extraction method using pyramid
histogram of orientation gradients for smile recognition. In: 16th IEEE Int. Conf. on Image
Processing, pp. 3305–3308 (2009)
9. Shao, H., Wu, Y., Cui, W., Zhang, J.: Image retrieval based on MPEG-7 dominant color
descriptor. In: 9th Int. Conf. for Young Scientist, pp. 753–757 (2008)
10. Yang, N., Chang, W., Kuo, C., Li, T.: A fast MPEG-7 dominant color extraction with new
similarity measure for image retrieval. J. of Vis. Commum. & Image Representation 19,
92–105 (2008)
11. Johnson, G., Song, X., Montag, E., Fairchild, M.: Derivation of a color space for image
color difference measurement. Color Research and Appl. 5, 387–400 (2010)
12. Talib, A., Mahmuddin, M., Husni, H., Loay, E.G.: A weighted dominant color descriptor
for content-based image retrieval. J. of Vis. Commun. & Image Representation 24,
345–360 (2013)
13. Swain, M., Ballard, D.: Color indexing. Int. J. of Computer Vision 7, 11–32 (1991)
14. Fierro, A., Nakano, M., Perez, H., Cedillo, M., Garcia, F.: An efficient color descriptor
based on global and local color features for image retrieval. In: Int. Conf. on Elect. Eng.
Comput. Science and Automatic Control, pp. 233–238 (2013)
15. Huang, J., Kumar, S., Mitra, M., Zhu, W., Zabih, R.: Image indexing using color correlogram. In: Conf. on Computer Vision Pattern Recognition, pp. 762–768 (1997)
16. Bleschke, M., Madonski, R., Rudnicki, R.: Image retrieval system based on combined
MPEG-7 texture and color descriptors. In: Int. Conf. Mixed Design of Integrated Circuit
and Systems, pp. 635–639 (2009)
17. Wong, K., Po, L., Cheung, K.: Dominant color structure descriptor for image retrieval.
IEEE Trans. on PAMI 32, 1259–1270 (2010)
18. Yap, P., Jiang, X., Kot, C.: Two-dimensional polar harmonic transform for invariant image
representation. IEEE Trans. on PAMI 32, 1259–1270 (2010)
19. Li, J., Wang, J.: Automatic linguistic indexing of pictures by a statistical modeling approach. IEEE Trans on PAMI 25, 1075–1088 (2003)
20. Wang, J., Li, J., Wiederhold, G.: SIMPLIcity semantic-sensitive integrated matching for
picture libraries. IEEE Trans. on PAMI 23, 947–964 (2001)
21. Kasutani, E., Yamada, A.: The MPEG-7 color layout descriptor: A compact image feature
description for high-speed image/video segment retrieval. In: Int. Conf. on Image
Processing, pp. 674–667 (2001)